Replace mrb_assert with mrb_ensure_*_type for VM opcodes that require
specific types:
- OP_ARYCAT: mrb_ensure_array_type
- OP_ARYPUSH: mrb_ensure_array_type
- OP_ASET: mrb_ensure_array_type (also fixed: was checking wrong register)
- OP_INTERN: mrb_ensure_string_type
- OP_HASHCAT: mrb_ensure_hash_type
These checks catch codegen bugs with clear error messages in both
debug and release builds.
Co-authored-by: Claude <noreply@anthropic.com>
Replace mrb_assert with mrb_ensure_string_type to catch codegen bugs
even in release builds. This prevents null-dereference crashes when
OP_STRCAT receives a non-string first operand due to compiler bugs.
Consistent with OP_HASH which uses mrb_ensure_hash_type for similar
type safety.
Co-authored-by: Claude <noreply@anthropic.com>
After pattern matching code generation, the sp (stack pointer) must
be restored to match the success path value. The failure path (after
RAISEIF) left sp in a different state, causing incorrect register
allocation in subsequent code like string interpolation.
This caused OP_STRCAT to use the wrong register, leading to
null-dereference when trying to modify a non-string value as a string.
Co-authored-by: Claude <noreply@anthropic.com>
The key variable in insertion_sort temporarily holds an array element
that's been removed from its slot during the sorting process. When
sort_cmp yields to a block that triggers GC, key wasn't protected
and could be collected.
Use arena save/restore around the loop to avoid arena overflow for
large arrays.
Test case from oss-fuzz: sort! with block containing rescue.
Co-authored-by: Claude <noreply@anthropic.com>
The => pattern matching codegen was doing push() after RAISEIF, even though
RAISEIF never returns. This caused sp to be off by 1 when success and failure
paths joined, resulting in wrong register allocation for subsequent operations.
For string interpolation like "#{ expr => pattern rescue body }", the base
string would be at R2 but STRCAT would incorrectly use R3, causing memory
corruption and crashes.
Test case: %{#{.=>.,. rescue def .()end}} (from oss-fuzz)
Co-authored-by: Claude <noreply@anthropic.com>
rational_new_b() expects both arguments to be bigints, but rational_new_f()
was passing an integer value for the numerator when the exponent was negative.
This caused a segfault in mrb_bint_reduce() which called RBIGINT() on the
integer value.
Test case: 5r**-92 (from oss-fuzz)
Co-authored-by: Claude <noreply@anthropic.com>
mruby-sleep: declare slp_tm before #ifdef _WIN32 block to fix
undeclared variable error in non-Windows branch.
hal-posix-socket: use #if defined(HAVE_SA_LEN) && HAVE_SA_LEN instead
of #ifdef HAVE_SA_LEN, since mruby-socket defines HAVE_SA_LEN to 0 on
non-BSD platforms.
Co-authored-by: Claude <noreply@anthropic.com>
Gems like mruby-task add preprocessor defines (MRB_USE_TASK_SCHEDULER)
that affect mrb_state structure. The amalgamation generator now detects
these defines from the build configuration and adds them at the top of
mruby.h before struct definitions are encountered.
Supported define patterns: MRB_USE_*, MRB_UTF8_*, HAVE_MRUBY_*
Co-authored-by: Claude <noreply@anthropic.com>
Add ability to generate combined mruby.h and mruby.c files for
single-file embedding, similar to SQLite's amalgamation.
Usage: rake amalgam
Output: build/<target>/amalgam/mruby.{h,c}
Features:
- Headers concatenated in dependency order with guards stripped
- Sources concatenated with proper ordering (core, gems, mrblib)
- X-macro headers (ops.h) inlined at each include point
- Local includes automatically inlined
- Handles both src/ and core/ gem directory conventions
Co-authored-by: Claude <noreply@anthropic.com>
Add #undef lesser after last usage to prevent macro redefinition
warnings when files are amalgamated into a single translation unit.
Co-authored-by: Claude <noreply@anthropic.com>
Remove unused mrb_stat typedef from file.c that conflicted with the
mrb_stat() function in file_test.c when compiled as a single
translation unit.
Fix convert_stat() in hal-posix-io to handle st_atime macro correctly
in both normal and amalgamated builds by extracting time values before
undefining the macros.
Co-authored-by: Claude <noreply@anthropic.com>
When parsing malformed input with many syntax errors (e.g., via eval
with a long garbage string), the parser would continue until the end
of input, causing long execution times.
Add an early termination check in the lexer that returns EOF once
the error count exceeds 10 (same as error_buffer size). This prevents
DoS from inputs like eval("garbage" * 1000).
Co-authored-by: Claude <noreply@anthropic.com>
mpz_div_2exp() was calling mpz_init_heap() on output parameter z
without first freeing z's existing memory. When called from
mpz_barrett_reduce() with pre-allocated temporaries, this caused
memory leaks.
Add mpz_clear(ctx, z) before mpz_init_heap() in both affected code
paths, matching the pattern already used in mpz_mod_2exp().
Fixes ClusterFuzz issue detected with input "8.pow 7*2515881+186,8 ^4>>-509".
Co-authored-by: Claude <noreply@anthropic.com>
Use self.begin/self.end instead of first/last to compute hash for
ranges. The first/last methods raise RangeError for endless/beginless
ranges, but the internal begin/end accessors return nil safely.
Co-authored-by: Claude <noreply@anthropic.com>
Support negative modulus in Integer#pow(exp, mod) with proper Ruby
semantics. Previously, negative modulus caused an infinite loop in
Barrett reduction. Now:
- Use absolute value of modulus for computation
- Apply signed modulo adjustment (result + m for non-zero result
when m is negative)
- Add early return for zero base with positive exponent (0^n = 0)
Co-authored-by: Claude <noreply@anthropic.com>
When right-shifting by more bits than the number contains, the loop
condition `i < x->sz - digs` would underflow (since size_t is unsigned),
causing out-of-bounds memory access.
Fixed by checking if digs >= x->sz upfront and returning zero in that
case, since shifting right by more bits than the number has always
yields zero.
Discovered via ClusterFuzz with input "7<<78<<-772".
Co-authored-by: Claude <noreply@anthropic.com>
In rational_new_f(), the code performed ((mrb_int)1)<<exp without
checking if exp >= MRB_INT_BIT. Shifting by a value >= bit width
is undefined behavior in C.
Also fixed the negative exponent case which incorrectly used
deno >>= exp (right-shift by negative is UB). The correct logic
is deno <<= -exp to multiply denominator by 2^(-exp).
Both cases now check for overflow before shifting and fall back
to bigint operations when necessary.
Discovered via ClusterFuzz with input "92r**11".
Co-authored-by: Claude <noreply@anthropic.com>
Fixed three bugs that caused infinite loops in GCD calculations:
1. mpz_set_int() didn't shrink sz when setting a smaller value.
mpz_realloc() only grows allocations, so setting a 1-limb value
to an mpz_t with sz=3 would leave sz=3, breaking algorithms
that depend on correct sz values.
2. mpz_set_uint64() had the same issue.
3. mpz_gcd() used mpz_init_set() which preserves the sign.
GCD should work with absolute values since gcd(a,b) = gcd(|a|,|b|).
With negative inputs, the sign would oscillate during mod operations,
preventing the Euclidean algorithm from converging.
4. mpz_div_2exp() when e==0 and z==x would corrupt data by calling
mpz_init_heap() which overwrites z->p before copying from x.
These bugs were discovered via ClusterFuzz with complex rational
number calculations.
Co-authored-by: Claude <noreply@anthropic.com>
Fix two functions that could create bigints with sn != 0 but value of 0:
- mpz_mod_limb: single-limb case set r->sn = x->sn even when result was 0
- mpz_mul_2exp: set z->sn = sn unconditionally after zero-producing ops
This inconsistent state caused GCD loop (!zero_p(&b)) to continue with
a zero divisor, eventually causing FPE in mpz_mod_limb with m = 0.
Co-authored-by: Claude <noreply@anthropic.com>
When generating code for pattern matching with potential failures, the
success and failure paths both need to pop the matched value. At
runtime, only one path executes. But during codegen, both pop() calls
affected the compile-time stack pointer (cursp), corrupting register
allocation and causing heap-buffer-overflow when accessing symbol
tables with wrong indices.
Fix by saving/restoring the stack pointer around the branch point, so
each path correctly tracks the stack state independently.
Co-authored-by: Claude <noreply@anthropic.com>
Pattern matching expressions were not pushing a result value in several
code paths when used in value context (e.g., string interpolation).
This caused crashes when the result was expected on the stack.
Fix all code paths in NODE_MATCH_PAT to push the appropriate value:
- 'in' pattern returns true/false
- '=>' pattern returns nil (matches CRuby behavior)
Co-authored-by: Claude <noreply@anthropic.com>
The direct literal generation optimization for parallel assignment was
using the RHS count as the loop bound but only filling registers for
LHS variables. When RHS has more elements than LHS (e.g., `a,=1,2`),
this caused uninitialized register indices to be used, generating
garbage opcodes that crashed the VM.
Fix by counting LHS variables and only applying the optimization when
LHS and RHS counts match exactly.
Co-authored-by: Claude <noreply@anthropic.com>
Add precedence declarations to resolve the ambiguity between:
- One-line pattern match: `arg in pattern`
- Case/in clause: `case expr; in pattern; end`
When seeing `arg in`, the parser should shift to parse `arg in pattern`
as a complete expression (matching CRuby behavior), not reduce `arg`
to start a case clause.
Changes:
- Add keyword_in to %nonassoc precedence declarations
- Add %prec tLOWEST to the plain `arg` reduction rule
This eliminates all bison shift-reduce conflicts (was 2, now 0).
Co-authored-by: Claude <noreply@anthropic.com>
Remove non-standard `symbol tASSOC p_as` rule from hash pattern
elements. This rule conflicted with the as-pattern rule and caused
`:foo => x` to be incorrectly parsed as a hash pattern instead of
an as-pattern.
CRuby only supports label syntax (foo:) for hash pattern keys,
not hashrocket syntax (:foo =>). This change aligns mruby with
CRuby behavior and reduces bison shift-reduce conflicts from 2 to 1.
Before: `case :foo; in :foo => x; end` raised NoMethodError
After: `case :foo; in :foo => x; end` binds x to :foo
Co-authored-by: Claude <noreply@anthropic.com>
The p_value grammar rule passed raw tSTRING token (a (len . str) cons
cell) directly to new_pat_value() without wrapping it as a proper AST
node. When codegen processed this malformed node, it read the length
field as the node type, causing misinterpretation and crash.
Wrap tSTRING with new_str(p, list1($1)) to create a proper NODE_STR,
consistent with how the primary:string rule handles strings.
Found by ClusterFuzz (oss-fuzz/mruby_fuzzer).
Co-authored-by: Claude <noreply@anthropic.com>
When rand is called with a range exceeding UINT32_MAX (e.g.,
rand(2..4294967297)), the span value could overflow when cast
to uint32_t, causing division by zero in the modulo operation.
Add 64-bit path for MRB_INT64 builds that combines two 32-bit
randoms when the range exceeds 32 bits.
Found by ClusterFuzz (oss-fuzz/mruby_fuzzer).
Co-authored-by: Claude <noreply@anthropic.com>
Replace for..in loops with while loops to avoid closure overhead.
Cache constants in local variables to avoid repeated lookups.
59% faster (7.4s -> 3.0s).
Co-authored-by: Claude <noreply@anthropic.com>
Optimized to avoid Math.sqrt by squaring the threshold
(sqrt(x) < 1000 => x < 1000000) and caching zr*zr/zi*zi
to avoid redundant computation. 29% faster than naive version.
Co-authored-by: Claude <noreply@anthropic.com>
Add syntax highlighting to mirb's multi-line editor with support for:
- keywords (def, if, class, end, etc.) in magenta
- strings ("...", '...', %q{...}) in green
- comments (#...) in gray
- numbers (42, 3.14, 0xff) in cyan
- symbols (:foo) in yellow
- constants (Array, Foo) in bold yellow
- instance variables (@var) in blue
- global variables ($var) in bold blue
Features:
- auto-detects light/dark theme via COLORFGBG env var
- MIRB_THEME=light/dark for explicit override
- method calls like obj.class correctly not highlighted as keywords
- enabled automatically when terminal supports color
Co-authored-by: Claude <noreply@anthropic.com>
Estimate initial buffer size based on format string to reduce
reallocations. The new formula uses format string length plus
120 bytes base, plus 24 bytes per format specifier, capped at 4096.
This reduces reallocations by ~60% in typical use cases and
improves performance by 2-21% depending on output size.
Co-authored-by: Claude <noreply@anthropic.com>
Ran `pre-commit run --all-files --hook-stage manual` and this ran prettier.
We had a Markdown table reformated and an backslash escape added.
A link was also fixed.
Tested both the standard and manual hooks pass
Use _WIN32 instead of _MSC_VER to provide strndup implementation
for all Windows compilers including MinGW/MSYS.
Co-authored-by: Claude <noreply@anthropic.com>
- Define strdup as _strdup on MSVC to avoid deprecation warning
- Add strndup implementation for Windows (not available in MSVC)
Co-authored-by: Claude <noreply@anthropic.com>
- rename NEWS to NEWS.md with markdown format
- document pattern matching (case/in) feature
- document new gems (mruby-task, mruby-benchmark, mruby-strftime)
- document mirb improvements
- document HAL platform abstraction
- document C API changes
- list fixed GitHub issues
- list 101 merged pull requests from community contributors
- list security fixes
Co-authored-by: Claude <noreply@anthropic.com>
Array#find is an optimized version of Enumerable#find for arrays,
using direct index access instead of each iterator.
Array#rfind finds from the end of the array, returning the first
match when scanning backwards.
Both methods support the ifnone parameter for default values.
Co-authored-by: Claude <noreply@anthropic.com>
Add notes section explaining the optimization behavior:
- Which functions are used for direct access
- When fallback to method dispatch occurs
- Why subclasses can override []/[]=
Co-authored-by: Claude <noreply@anthropic.com>
Replace mrb_obj_class() with direct mrb_obj_ptr(va)->c access:
- Skips unnecessary mrb_immediate_p() check (these types are never immediate)
- Skips mrb_class_real() traversal for singleton classes
- Objects with singleton methods now fall back to method dispatch
(correct behavior since they might have overridden []/[]=)
Co-authored-by: Claude <noreply@anthropic.com>
Add inline optimizations for Array#[]= and Hash#[]= in OP_SETIDX,
matching the pattern established for OP_GETIDX:
- Array class: use mrb_ary_set() directly (integer index only)
- Hash class: use mrb_hash_set() directly
- Subclasses: fall back to method dispatch (can override []=)
- String: unchanged (complex 2-3 argument signature)
Co-authored-by: Claude <noreply@anthropic.com>
John Bampton